Internal accountability note, prepared October 5 on the October 5 settled screen (desk date = SPY's October 5 close). This is the Loop 4 companion to the Monday flagship. It is the one place the desk reads its own books, so paper ledgers and session logs are cited below; the real-money accounts were not read, and none of these judgments enter the public brief. Nothing was fetched or backtested for this note. Prices and returns are fields the engine computed in research/market-engine/data/summaries/*.json (October 5 settled close unless dated) or figures printed by the week's briefs, scans and call projection; none was computed by hand.

The short version

• The screen: miss-screen read 1,254 symbols and found 41 movers at +30% or more over 30 days with $10 million or more of daily dollar volume (one stale mover skipped). Four were held or claimed; 37 were not. They form two own-or-decline clusters (AI bottlenecks / optics / co-packaged optics, 14 names; biotech tools, 9) and a 14-name tail.
• It is mostly last week's list. 14 of the September 28 autopsy's 20 names are back, unclaimed. DOT-USD, SDGR and SMTC have been on every screen since September 21. TXG is in its seventh autopsy since July 5.
• It is not a coverage miss. All 37 sit on at least one watchlist. 31 printed on the October 2 monster scan and five more were named in a brief or scan that week. Only Lasertec (6920.T) surfaced nowhere. Both clusters were read during the week and passed by written rules: "buy flags, not gaps" for optics and chip tools, and the biotech campaign's "RSI > 65 without a prior reset is a documented pass."
• The decision step is what failed. The September 28 autopsy asked for one grouped own-or-decline pass over crypto, semis, biotech and freight. No decision is on file, so the screen raised all four again. The screen also can't remember a decision: it reads the emitter's folded verdict line as the literal text >-.
• Called right: rates as the week's dividing line. Monday's AI selloff read as price pressure, not a spending reversal. Neoclouds sorted by trend. Micron's pivot waited for. Accenture's pop not chased. Seven of the nine pass calls that reached review this week trailed SPY at 30 days.
• Wrong or lapsed: the July AXTI pass, +43.6 points over SPY at 30 days, reached its October 1 review date and nobody reviewed it. 54 of 80 open desk calls are past their review date.
• The books did what the research said: no chasing. One book queued a build-out position, and it was a basket. Two books were blocked by human mandate questions open since September 25, one of them on SOXL in a week the semiconductor fund rose 3.96%. A quality book honored its own pre-written SLB exit two sessions late.

What the screen says

deno task market -- miss-screen, run October 5. MISS means no book holds the symbol and no active or monitoring campaign lists it in ACTIVE.json key_tickers. On its own it does not show that the desk failed to notice a move, passed an entry, or had an executable trigger at the prior close. Each row answers the operating book's three questions: in our data? which thesis should own it? what trigger at T−1? "(signal row)" marks a generated line in a brief, not desk-written prose.

Cluster 1 — AI bottlenecks, optics, co-packaged optics (14 MISS; 4 repeat from September 28)

The screen prints "$345,473M aggregate dvol" for this cluster. That figure is the wrong basis. The four Asian listings contribute 96.0% of it, and their volume is in yen and Taiwan dollars summed as if it were dollars. The eight US-listed names total $13.7 billion.

Disposition. A conviction miss by design. The briefs named the group daily: "Extended: buy flags, not gaps" (10-01); "nothing new to buy at Friday's prices; set the levels" (10-02). The real ownership gap is narrower than 14 names. optical-supercycle claims COHR, LITE, CIEN, FN and CRDO. Its own October 2 vertical argues about MXL, Semtech and AXT, but none of them is in key_tickers, so the screen keeps calling them unclaimed.

Cluster 2 — biotech tools and genomics (9 MISS, $2.4 billion aggregate dollar volume; 3 repeat from September 28)

Context first: over the same 30 days IBB is −1.87% and XBI −4.65%. This is a run in small and mid-cap genomics, proteomics and AI-drug-discovery tools inside a soft sector. It is not a biotech rally.

Disposition. This is not coverage and not a silent pass. On 07-26, after three autopsy carries, the campaign wrote a mid-cap entry leg:

• arm on a pullback reset (TXG or CERT RSI < 45);
• convert on the resume (RSI back over 55, price over the 20-day);
• "RSI > 65 without a prior reset is a documented pass, not an entry."

TXG never reset. Every week since has been that documented pass, and the screen has re-raised it every week. Three things make it look worse than it is:

1. The registry has drifted from the README. The campaign README lists eight anchors (TMO, DHR, IQV, MEDP, RGEN, TWST, SDGR, ABCL). ACTIVE.json carries five. SDGR's three straight MISS lines come from that gap.
2. The machine reads a sequence as two separate clauses. trigger-check prints "TXG rsi > 55 … FIRED · STANDING 20+d" while the arm clause "TXG rsi < 45" is DORMANT. The rule is arm-then-convert. Read clause by clause, its "fired" line has meant nothing for 20 sessions.
3. The campaign record is quiet. The log's last dated entry is 07-07. The README was last refreshed 09-06.

The campaign's own escalation gate is now close. TMO crossed RSI 60 on the 10-05 bar; DHR (57.5) and MEDP (56.1) are armed. Once all three clear 60, the July rule enters the mid-caps on the next confirmed setup, and the leaders are still extended.

The unclustered tail (14)

What the desk called right and wrong

Right:

• Rates. The 09-30 climate read and briefs ("Rates are the weather") framed the week as a rates shock under a quiet index. The week ended that way: S&P 500 fund −0.22%, semiconductor fund +3.96%, and every bond and credit fund below RSI 25 (10-02 brief).
• Monday's AI selloff. The 09-28 brief kept it "classified as price pressure rather than a proved spending reversal". By Friday, equipment and test had closed 13 for 13 for a second straight Friday, and optics 25 of 28.
• Neoclouds sorted by trend. 09-29: "Nebius is the only pure-play whose long trend survived September." For the week: Nebius +2.31%, Cipher −11.39%, Hut 8 −7.43%, IREN −5.37%, Cerebras −19.5%.
• Micron. 09-30: "the trade is a settled close above $1,108.72, not an after-hours quote". It never closed there: −2.05% Friday, $1,063.89 on 10-05.
• Accenture. On 10-01, at +15.78%, the desk called it "the pop; the entry is a flag that holds the 200-day (about $196.29)". It fell 6.31% Friday and closed $195.05 on 10-05, under that line.
• Passes at 30 days. Nine pass calls reached their review date this week. Seven trailed SPY over their 30-day window (name minus SPY, percentage points): ITA −12.18, KTOS −23.29, RCAT −23.76, CIFR −25.11, CORZ −14.34, CRWV −3.00, HUT −20.25. ORCL was +1.90 and AXTI +43.60. The call ledger's 30-day pass channel agrees: 6 of 27 passes missed a runner, mean forward excess −10.63. That is the opposite of the 09-08 finding that "the desk's passes are wrong three times in four".
• The 09-05 supplier buys. The industry report's nine buys were superseded the next day by the Campaign 1 entry map. They have now completed their 30-day window. Six are ahead of SPY (excess points): AMAT +13.54, COHR +9.36, LRCX +6.76, MRVL +19.18, SOXX +10.41, TSM +9.50. Three are behind: GLW −5.13, VRT −13.95, WDC −8.63. CALL-LESSONS re-reads on 10-06 and still says buys "looked uniformly bad" at 30 days; these grades belong in that re-read.

Wrong, or let lapse:

• AXTI, above. It is the pass with the most concrete written reopen condition, and nothing ever checked that condition.
• The two July "too extended" passes show their cost after 30 days. AXTI is +44.15% raw since the call. OKTA was a good pass at 30 days (−2.68) and is +48.73% raw since its 07-08 call; its 90-day window is still open.
• SNDK. The 09-07 watch's $1,740 line closed above on 10-01 ($1,787.69) with no flag recorded, and its 09-30 review date passed. It was $1,704.16 on 10-05. Probably not an entry, since the call wants a new flag first, but the call's lifecycle lapsed.
• Review dates. On 10-02, 54 of 80 open desk calls were past their review date. Fifteen review dates fell inside the week and none was closed.
• The 09-25 weekly's shelves. Four of its watch calls were invalidated within four sessions: CRWV (09-28 close), VSAT (09-29), AKAM (09-30), AVGO (10-01). Watch calls are unscored; this is an observation about where the shelves sat, not a grade.

What the books did versus what the research said

The research said don't chase the build-out, and the books didn't.

• No book bought a chip, optics or biotech-tools name all week.
• Eleven Tier B books held sessions on the 10-02 close and none bought.
• The one build-out position queued is swing-kid's photonics basket, LAZR: 3% of the book, a 20-day breakout on 2.23x volume, for Monday's noon fill. A basket sidesteps the single-name stretch the briefs warned about.
• The aggressive book is +10.80% since inception against the S&P 500 fund's +16.21% (10-02 report card).

The week's fills were mostly exits:

• degen: SPXL sold 09-29 (trend died, $283.20); TQQQ sold 10-01 (10-day time stop, $78.73, +1.4%).
• yolo: −20% hard stops on CRS (09-28, $385.30) and NKE (10-01, $35.15, before the post-report after-hours quote of $32.25).
• momentum-trader: DAR stopped out (09-29, −8.2%, $60.32).
• claude-trader: ABBV hit its +30% target (09-28, $266.28); UNH bought (09-28, $374.97), mirrored in the hold and trail arms.
• Trail arms: ARM sold at $283.33 on 09-28, the day ARM fell 8.7%; ARM closed $302.90 on 10-05. bench-signals-trail also sold RBLX and NTR.
• dca-larry made its October buys; autotrade held everything.

Three findings:

1. Human questions blocked books on the exact tape the research named. Three mandate questions have sat in TASKS-ALGO.md since 09-25:

   • degen's SOXL authority (SOXX or SMH). UNDECIDED every night 09-29 → 10-02, while the 10-01 session said "semis pass every price test". SOXL is +40.07% over 30 days and +15.45% over 7, and it is on tonight's screen.
   • mention-scalper's Friday rule. The code fails every Friday; the playbook says a Friday session may nominate. This was the third Friday the worker named it, and it held four clean breakouts (PENG, VSH, IBRX, IMOS). IMOS is on the screen.
   • value-vulture's missing Rule 2 calculator. General Dynamics cleared the quality bar again and stopped there.

   Each worker correctly refused the reading that would have let it trade. The cost belongs to the open question, not the worker.

2. A pre-written exit fired late because it lived in prose. When claude-trader bought SLB on 09-18 it wrote that "a confirmed close below the 200-day is a thesis-broken exit". SLB crossed on 09-29 and confirmed on 09-30. The book honored the exit on 10-02, "two sessions late", for a Monday fill. Its own note: pre-registered price lines "should live where the nightly exit engine evaluates them". This is the 09-21 lesson ("a condition written in a brief is a trigger-check row, or it doesn't count") showing up on the exit side.

3. The index book has no research to admit against. After the time stop cut TQQQ, the box nominated TQQQ ENTER on 10-01 and again on 10-02. degen passed both times: no current artifact makes the Nasdaq-100 case, and the three campaigns that covered mega-cap tech all closed on 09-06. QQQ closed $756.20 on 10-05, 0.1% under its high; TQQQ closed $83.12 against the $78.73 exit. The gate worked as written. The open question is whether a book that trades index wrappers can run on a corpus that only writes sector-supplier theses. The existing thesis-supply row covers value-vulture and claude-trader, not degen.

Exits that sell a one-day drop inside an intact trend keep showing up. This week it was the ARM trail sale; on 09-21 it was MU's stop at its exact low. In the call ledger's 30-day stop-loss channel, 17 of 28 stopped names went on to outperform (mean forward excess +5.03). The sample is small; noted, not ruled on.

Defects. Nine escape rows were filed during the week (D192–D200). Each was caught by an agent, the operator, a cold audit or the cold grader; none by a gate, and none is recorded as reaching a ledger. One reached a box: D199's misread Blackstone insider leg nominated BX ENTER in yolo and yolo-insider-probe, and the workers refused it by hand.

Lessons that repeat

New this week: human mandate questions that sat for ten days are now a direct cause of book inaction.

Proposed (not filed; the user's call)

Three changes, smallest first. None adds a gate.

1. Decide each cluster once, in a form the screen reads. This note carries a short single-line verdict. Fix miss-screen's prior-verdict parse (parsePriorMissAutopsy, cli/synthesis/miss-screen.ts:329), which reads the emitter's folded verdict: >- as the verdict. Then rule on the standing groups:
   • Biotech: decline the small-cap squeeze. PACB, NAUT and QSI are not our game (the campaign itself called NAUT and QSI "not core"). TXG, GRAL and ILMN are documented passes under the July rule. Restore the README's anchors (SDGR, TWST, ABCL) to ACTIVE.json so the registry matches the campaign.
   • Optics and co-packaged optics: own the US-listed component layer inside optical-supercycle (MXL, SMTC, AXTI, VIAV, the names its 10-02 vertical argues about), using the vertical's flag-and-retest entry. Decline the foreign primaries under the research-only rule.
   • Crypto and BWET: own or decline; both are owed for a third week. BWET also needs its tradability answer.
   • SEI, COHU, AMD: name an owner or decline.
2. Machine-watch every written condition that carries a number, through the existing trigger-check/call-check path. That covers a pass's reopen line, a book's pre-registered exit, a campaign's arm-then-convert sequence, and a call's review date (a lapsed review prints that night). These are items 2–3 from 09-21, still unbuilt. This week they would have surfaced AXTI on 10-01, SLB on 09-30, and the meaningless TXG "FIRED".
3. Answer the three parked mandate questions in one sitting: degen's SOXX-or-SMH authority, mention-scalper's Friday rule, and value-vulture's Rule 2 calculator. All three sit above their books' Evolved Rules line, so only the user can settle them, and each one blocked a book this week.

Open checks that fell out of this note (task rows to file, not done here):

• miss-screen folded-verdict parse.
• miss-screen aggregate dollar volume, and the summaries' dollarVolume field, sum local-currency volume for foreign listings as dollars. This is a currency-basis error, and it also means the $10M floor is applied in local currency for those names.
• trigger-check evaluates sequential watch_for clauses independently.
• KOD's two contradictory cause statements (09-28 vs 10-02 Read).
• The biotech campaign README carries a "User holdings inside the lane" line: book information inside a research document.
• The SNDK, MXL and AXTI calls are past their review dates.

Sources and limits

• Screen receipts (October 5): miss-screen, trigger-check, call-ledger report (run during the October 5 EOD; outputs not persisted); join logic research/market-engine/cli/synthesis/miss-screen.ts.
• Prices: research/market-engine/data/summaries/*.json, October 5 settled close.
• Desk reads: research/issues/2026-09-28-market-brief.md through 2026-10-02-market-brief.md; research/issues/2026-10-02-vertical-optical-photonics-lasers-not-modules.md; research/investigations/2026-09-30-climate-read-rates-shock.md; research/scans/2026-10-02-{biotech-scan,optical-supply-chain,wfe-test-metrology,ai-infrastructure,monster-scan,crypto-scan}.md; research/targets/2026-09-23-sdgr-target-entry-zone.md, 2026-09-29-amd-target-entry-zone.md.
• Calls: research/api-data/desk-calls.json (generated 2026-10-06T00:39Z); research/workflow/CALL-LESSONS.md.
• Campaigns: research/perspectives/ACTIVE.json; research/perspectives/2026-06-04-biotech-capital-cycle/{README,log}.md; research/market-engine/watchlists/_securities.json.
• Books: research/trading/paper/*/PLAYBOOK.md sessions 09-28 → 10-02; research/trading/paper/*/LEDGER.json fills in that window; research/trading/paper/autotrade/runs.jsonl; ops/tasks/TASKS-ALGO.md (open mandate rows).
• Defects: DEFECTS.md D192–D200.
• Prior autopsies: 09-21, 09-28; earlier TXG carries in research/investigations/2026-07-{05,13,18,26}-*miss-autopsy*.md and 2026-08-25-miss-autopsy-week-ending-2026-08-21.md.
• Limits: no execution replay, raw OHLC read, company deep dive or news pull. Counterfactual P&L for blocked or passed entries is not computed. The six monster-scan-only tail names get visibility lines only.